Options
One-Class SVM Outlier Analysis
Contributor II
Hi Everyone,
I am trying to use One-Class SVM as an outlier analysis algorithm. I have successfully applied this using LibSVM using the command line version in linux and windows. I want to simply replicate the process in RM.
Currently the LibSVM operator does not seem to behave as the LibSVM command line programs when used in One-Class mode.
here is the goal take data with 50% normal sample sand 50% outliers all with the same label, and produce a one-class model that captures the normal samples and separates out the outliers. Again, this was possible with libsvm command line version. The confidence outputs of the LibSVM operator seem to correctly split the values however I cant get the predict confidence operator to split correctly.
Any ideas?
Thanks,
-Gagi
I am trying to use One-Class SVM as an outlier analysis algorithm. I have successfully applied this using LibSVM using the command line version in linux and windows. I want to simply replicate the process in RM.
Currently the LibSVM operator does not seem to behave as the LibSVM command line programs when used in One-Class mode.
here is the goal take data with 50% normal sample sand 50% outliers all with the same label, and produce a one-class model that captures the normal samples and separates out the outliers. Again, this was possible with libsvm command line version. The confidence outputs of the LibSVM operator seem to correctly split the values however I cant get the predict confidence operator to split correctly.
Any ideas?
Thanks,
-Gagi
0
Answers
http://rapid-i.com/rapidforum/index.php/topic,1746.msg6795.html